Effective Scheduling Algorithm for Workload Forecasting in Fog Environment Utilizing
Dual Interactive Wasserstein Generative Adversarial Network
SuggalaRavi Kumar1,*
MSuma Bharathi.1
KumarP.L.V.D. Ravi1
KumarNVS. Pavan2
-
( Assistant Professor, Department of Information Technology, Shri Vishnu Engineering
College for women (A), Bhimavaram, Andhra Pradesh, India ravikumars, suma.thota,
plvdravikumarit}@svecw.edu.in)
-
( Associate Professor, Department of Computer Science and Engineering, Koneru Lakshmaiah
Education Foundation, Vaddeswaram, AP, India nvspavankumar@kluniversity.in )
Copyright © The Institute of Electronics and Information Engineers(IEIE)
Keywords
Fog computing, Improved dwarf mongoose optimization algorithm, Fog resource monitor
1. Introduction
Fog computing is a hybrid paradigm that emerged from recent advances in mainstream
parallel with distributed computing (PDC) technologies, such as cloud and edge computing.
One of the best solutions currently available for leveraging distant computational
cloud resources and performing calculations near the network edge is fog computing
architectures [1]. Such settings leverage containers for lightweight, flexible, and fine-grained resource
sharing amongst several fog devices. On the other hand, task scheduling is challenging
in heterogeneous fog environments with unpredictable workloads [2]. Due to the shift in workloads to AI, ML, and DL, modern users also require extremely
low reaction times and energy usage. The resource heterogeneity makes assigning tasks
in these fog environments more complicated because different tasks have different
preferences for the best cost-performance trade-off.
Existing Solutions: Recent works employ dynamic scheduling approaches that adjust
to varying infrastructure conditions in real time to address the difficulties of volatile
tasks scheduling in diverse fog environments [3]. A lot of schedulers adjust their estimate of the "expected reward" for dynamic QoS
improvement using techniques like reinforcement learning. Because of their slow learning,
long scheduling delays, and brittle modelling assumptions, they not necessarily appropriate
for high non-stationary applications. Several approaches use virtual queues for scheduling
in constrained contexts and combine maximum weight-based tactics with belief-based
exploration to get around these issues. Recent research utilizing a deterministic
deep surrogate model1 and gradient-based optimization has been shown to outperform
such methods. Deep neural networks (DNN) [4] deterministic surrogate models, and gradient-base optimization enable less energy
consume and quick response time. It is a result of its rapid real-time adaptation
to various scenarios. Such techniques might not work well in situations unfamiliar
to the model during training because they do not include state-space exploration or
uncertainty modeling. Ignoring the curvature information of QoS optimization surfaces
can result in these methods being trapped at local optima, because these surfaces
are highly non-convex [5].
New Insights: To navigate the difficulties of unobserved settings, it is imperative
to take uncertainty in the model predictions into account. Uncertainty generally results
from various factors, including approximations in the model, inaccurate measurements,
and parameter variations over time. In particular, a deterministic mode does not provide
the entire distribution because of its unknown nature or the inherent randomness of
certain factors, such as network delay, temperature, hardware defects. Therefore,
an Effective scheduling approach for workload forecasting in a fog environment utilizing
DIWGAN is proposed.
Talaat et al. [6] suggested a Scheduling approach for load balancing in fog computing utilizing CNN-MPSO,
which attains a higher make span and lower LBL. Tuli et al. [7] Suggested GOSH: TS-GOSH-FCE, which provides a higher total cost and lower ARU. Teoh
et al. [8] suggested IoT with fog computing depending on predictive care methods for effectively
managing assets in Industry4.0 utilizing machine learning. It attains higher ARU and
lower LBL. Sharma et al. [9] suggested Two-Stage Optimum Task Scheduling for a Smart Home Environment Utilizing
Fog Computing. This system provides higher total cost and lower makes span.
The primary contributions of this paper are abridged below:
·EDLB using DIWGAN is proposed, which comprises three modules: (i) fog resource monitor,
(ii) DIWGAN base classifier, (iii) optimized dynamic scheduler.
·When EDLB is compared with various LB approaches from the past, it achieves high
resource utilization while reducing response time.
·Thus, it is a proficient way to guarantee consistent service. As a result, EDLB is
straightforward and effective in real-time fog computing structures, such as the healthcare
system.
·LB for FC has introduced several methods, but they all have numerous drawbacks. EDLB
overcomes these restrictions and performs well in various situations.
Remaining paper is arranged as: division 2 illustrates the proposed method, division
3 proves the results, division 4 concludes this paper.
2. Proposed Methodology
This segment discusses task scheduling using DIWGAN optimized with Improved Dwarf
Mongoose Optimisation Algorithm (ESA-WF-FE-DIWGAN) [10]. Fig. 1 portrays the block diagram of ESA-WF-FE-DIWGAN task scheduling system. The comprehensive
description of every stage is deliberated beneath:
Fig. 1. ESA-WF-FE-DIWGAN task scheduling system.
2.1 Data Collection
Cloud-Fog computing is a hybrid computing paradigm that maximizes resource consumption
and enhances overall system performance by merging the power of cloud computing with
the proximity of fog computing [11]. Dynamic load balancing is a crucial part of this paradigm because it involves distributing
computational tasks and network traffic efficiently across fog and cloud nodes based
on workload variations and resource availability.
A Cloud-Fog Computing dataset used for load balancing typically consists of various
parameters and metrics that reflect the current state of the system. These parameters
include:
1. Workload characteristics: This includes the number and type of computational tasks,
their processing requirements, arrival rates, and deadlines. The workload characteristics
help evaluate the load on the system and determine the appropriate resource allocation.
2. Resource availability: This refers to the current status and capacity of fog and
cloud nodes. It includes CPU utilization, memory usage, network bandwidth, and power
consumption. These metrics are critical for assessing the available resources and
making load-balancing decisions.
3. Network conditions: The dataset may also include network-related parameters, such
as latency, bandwidth, and network traffic. This information helps determine the optimal
placement of tasks based on the network proximity and congestion levels.
The number of data samples in a Cloud-Fog Computing dataset can vary under particular
requirements and the complexity of the simulated environment. Typically, datasets
used for load-balancing experiments contain a significant number of samples to capture
different workload scenarios, resource states, and network conditions accurately.
The dataset may provide details about the characteristics and configurations of the
fog including cloud nodes. Fog nodes typically deployed at the network edge, near
to the data source, and offer low-latency processing capabilities. They can be heterogeneous
on the processing power, memory, and network connectivity. On the other hand, cloud
nodes are located in data centers and offer vast computational resources with higher
latency.
2.2 Fog Resource Monitor (FRM)
As shown in Table 1, this is in charge of observing every server resource, also recording the server
data in the FRT database. FRT is situated in FRM server. FRM server is a fog server.
Every fog server is categorized as appropriate or in appropriate as per the set of
rules depending on storage, computing, and RAM.
The given steps determine the stages of every server are
i. Compute average storage$(AVG-STR)$.
ii. Compute average computing $(AVG-CMT)$
iii. Compute average RAM size$(AVG-RAM)$.
iv. Applying the following set of rules: (a) If $STR\geq AVG\_ STR$then$STR=1,$ ELSE
IF $STR<AVG\_ STR$then$STR=0.$ (B) If $CMT\geq AVG\_ CMT$then$CMT=1,$ ELSE IF $CMT<AVG\_
CMT$then$CMT=0.$ (C) If $RAM\geq AVG\_ RAM$then$RAM=1,$ ELSE IF $RAM<AVG\_ RAM$then$RAM=0.$
STR, CMT, RAM are the input of DIWGAN to train the DIWGAN. The following section outlines
the training process and DIWGAN CNN.
Table 1. Fog resources table.
|
Si
|
Storage (GB)
|
Computing (MHZ)
|
RAM (GB)
|
Type
|
Status
|
|
S1
|
100
|
4000
|
7
|
Adequate, proper
|
Suitable
|
|
S2
|
90
|
5000
|
8
|
Proper, not adequate
|
Inappropriate
|
|
S3
|
200
|
2700
|
4
|
Adequate, not proper
|
Inappropriate
|
|
S4
|
40
|
5000
|
6
|
Adequate, not proper
|
Inappropriate
|
|
S5
|
130
|
5000
|
7
|
Adequate, proper
|
Suitable
|
|
S6
|
80
|
6000
|
7
|
Proper, inadequate
|
Not suitable
|
|
S7
|
120
|
7000
|
8
|
Adequate, proper
|
Suitable
|
|
|
…
|
…
|
…
|
…
|
…
|
|
Sn
|
…
|
…
|
…
|
…
|
…
|
2.3 Classification using DIWGAN
This section discusses the DIWGAN, which categorizes each fog server. The data were
sent to the Fog Resource Monitor (FRM), and each monitor synthesized one server. The
FRT communicates to each other for sharing data from diverse inputdata storage [12]. The data distributions from FRM and the features stored database are compared using
the Wasserstein distance to ensure FRM yielded accurate server information. The min–max
problem between FRM can be described in Eq. (1),
The initial two terms refers Wasserstein distance assessment;$X$denotes input data
from FRM;$F_{X-{Q_{x}}}$represents the collected database. $G$denotes the storage
and $R$is the computing power. FRT epitomizes penalty coefficient. The network structure
for the algorithm is provided as follows:
2.3.1 Generative Model
FRM consists of two paths they are contracting and an expansive path. By using an
expanding path to comprehend information exchange and material deterioration, two
monitors take features from the contracting path. The database fitters used to extract
the features were 32, 48, 68, and 152.
2.3.2 Selector
The distance between the database and the fog resource
monitor can be represented as Eq. (2)
where the subscript $l$represents the FRM. By comparing the values, the performance
of the databases was examined during each iteration. The FRT sends features to the
network, and it classifies the features with the efficiency of DIWGAN in the next
iteration.
2.3.3 Hybrid Loss Function
The data from FRT sends the features and is represented as Eq. (3),
where$h$represents the height;$b$is the width;$d$is the depth of the input sensor
data. The method to improve the performance is expressed as Eq. (4)
where$p$and$\,q$denote gradient descent directions for input data$P$ and $Q$. $\Gamma
$denotes the suitable and unsuitable cauterization, resulting in a more realistic
result that is expressed as Eq. (5),
The features are transmitted to DIWGAN, which is represented in Eq. (6),
The FRM can yield several databases by minimizing the features and classifying the
servers into suitable and unsuitable using DIWGAN. Therefore, the classified servers
are fed to the Task Scheduling process.
2.4 Task Scheduling using Improved Dwarf Mongoose Optimisation Algorithm
In this section, Task Scheduling using the Improved Dwarf Mongoose Optimisation approach
(IDMO) is discussed [13]. The Improved Dwarf Mongoose Optimization Algorithm (IDMOA) offers several advantages
for task scheduling and process assignment. IDMOA excels in assigning incoming processes
to the most suitable server. It optimizes the allocation process by considering various
factors, such as the server load, available resources, and processing capabilities.
This ensures the processes are distributed effectively and efficiently, improving
the overall system performance. IDMOA is highly scalable and suitable for large-scale
systems and complex task-scheduling scenarios. It can handle many processes and servers
while maintaining efficient process allocation. As the system grows, IDMOA can adapt
and optimize the process assignment to accommodate the increased workload. IDMOA maximizes
resource utilization through improved resource utilization considering the server
load and available resources. It avoids overloading servers and distributes processes
evenly across the available servers. This optimization results in the better utilization
of system resources, reducing the idle time and increasing the overall efficiency.
It exhibits robustness in handling dynamic environments and fluctuations in process
arrival rates. It can adapt to changing conditions and adjust the process assignment
strategy dynamically. This flexibility allows the algorithm to maintain optimal performance
even in scenarios with varying workloads and server availability. IDMOA aims to minimize
the response time of processes by efficiently assigning them to appropriate servers.
Considering factors, such as server proximity and workload reduces the time taken
to execute tasks. This leads to faster completion of processes and improved system
responsiveness. IDMOA can be customized and fine-tuned to suit specific system requirements
and objectives. The algorithm allows for the incorporation of different optimization
criteria, such as energy efficiency or load balancing, based on the specific needs
of the system. This flexibility enables the algorithm to adapt to diverse optimization
goals. The incoming process must be assigned to the most suitable server by the IDMOA.
When analyzing EDLB, including LB approaches, it may reduce the response time and
attain higher resource usage. Therefore, this is an excellent way to confirm the endless
service.
IDMOs are used for allocating the incoming tasks to a suitable server. An IDMO chooses
the appropriate server from the obtainable appropriate servers in the Fog Resource
Monitor.
2.4.1 Population Initialization
The IDMO populaces are initialized by a matrix of candidate dwarf mongooses $(Y)$that
can be represented in Eq. (7),
where $m$ is the number of dwarf mongooses. $y_{1,1}....y_{1,e-1}$ represent the storage,
computing, and RAM features from the server.
2.4.2 Alpha Group
Assign the incoming process that is selected as the alpha, which controls the chaos
and fits the database as expressed in Eqs. (8) and (9),
where$Y_{{_{j}}}$ and $Y_{i}$ are the selected dwarf mongooses.
2.4.3 The Babysitters
The dwarf mongooses (DM) replace the baby sitters, which means the information sent
to servers and select the best incoming task and is represented in Eqs. (10) and (11)
where the best incoming task is selected, and again, the process is repeated. Thus,
task scheduling is performed based on the IDMOA.
2.4.4 Termination
In this step, the task is scheduled with the help of IDMOA to iteratively repeat step
3 after selecting the process. Finally, the ESA-WF-FE-DIWGAN performs task scheduling
with a higher ARU with a lower make span.
3. Result and Discussion
The simulation outcomes of the proposed scheduling algorithm (SA) for workload forecasting
in FE utilizing the DIWGAN (ESA-WF) method is discussed. The simulation is done in
JAVA using iFogSim simulator tool on Windows 10OS. The proposed ESA-WF-FE-DIWGAN method
achieved feasible outcomes under several performance metrics, such as Makespan, Total
cost, ARU, and LBL. The obtained results of the proposed method ESA-WF-FE-DIWGAN were
analyzed with the existing methods, such as TS-LB-FE-CNN-MPSO and TS-GOSH-FCE, respectively.
3.1 Performance Measures
The performance metrics, such as Make span, Total Cost, ARU, and LBL, are examined.
3.1.1 Makespan
The total time requisite to accomplish all tasks is scaled using Eq. (12).
Here$CT$ denotes time $ti$ completes its processing.
3.1.2 Total Cost
Assume a task is completed in a cloud-fog system; Eq. (13) shows that processing, memory use, and bandwidth are charged minimally:
Here$CP$ denotes processing cost.
3.1.3 ARU
Every resource usage present in the ecology of fog and is represented in Eq. (14),
Here$BS$ denotes count of balancing$FS_{s}$; $OL$ represents count of overload$FS_{s}$;$FS_{s}$indicates
a count of obtainable fog servers.
3.1.4 LBL
LBL measures the load level. A count of overloaded fog servers is subdivided from
the percentage of balanced fog servers represented in Eq. (15):
3.2 Performance Analysis
Tables 2-5 presents the outputs of the proposed ESA-WF-FE-DIWGAN method. The performance metrics,
such as Makespan, Total cost, ARU, and LBL are analyzed. Here, the proposed ESA-WF-FE-DIWGAN
technique is compared with existing methods, such as TS-LB-FE-CNN-MPSO and TS-GOSH-FCE.
Efficient dynamic load balancing performance analyzed with existing approaches by
deeming the metrics. Table 2 shows a comparison of Make span.
The proposedESA-WF-FE-DIWGANmethod provides 32.21%, which was a 37.56% lower Make
span for50 tasks; 25.45%, which was a 45.89% lower Make span for 90 tasks; 56.34%,
which was a 21.23% lower Make span for 130 tasks; 25.45%, which was 45.89% lower Make
span for 150 tasks; evaluated as the existing methods TS-LB-FE-CNN-MPSO and TS-GOSH-FCE
Table 3 compares the total cost. Here, the proposedESA-WF-FE-DIWGANmethod provided32.21%,
which was a 37.56% lower entire cost for 50 tasks; 25.45%, which was 45.89% lesser
entire cost for 90 tasks; 56.34%, which was a 21.23% lowerentire cost for 130 tasks;
25.45%, which was a 45.89% lower Make span for 150 tasks; these were evaluated to
the existing methods TS-LB-FE-CNN-MPSO and TS-GOSH-FCE.
Table 4 compares the ARU. Here, the proposedESA-WF-FE-DIWGANmethod provides a 56.45% and
67.68% higher ARU for 50 tasks; 73.56%, which was 76.44% higher ARU for 90 tasks;
57.57%, which was 59.3% higher ARU for 130 tasks; 67.78%, which was 45.89% higher
ARU for 150 tasks; these were evaluated to the existing methods TS-LB-FE-CNN-MPSO
and TS-GOSH-FCE.
The proposed approach produced better performance than the existing algorithms as
greater ARU and higher LBL are attained, as listed in Tables 3 and 4. The experimental outcomes show that the proposed algorithm consumed less and
reduced the number of migrations.
Table 5 compares the LBL. Here, the proposedESA-WF-FE-DIWGANmethod provided a32.21%, which
was a 37.56% higher LBL for 50 tasks; 25.45%, which was a45.89% higher LBL for 90
tasks; 56.34%, which was a 21.23% higher LBL for 130 tasks; 25.45%, which was a 45.89%
higher LBL for 150 tasks; these were compared with the existing methods TS-LB-FE-CNN-MPSO
and TS-GOSH-FCE respectively.
Furthermore, the cost was paid for the processor, memory, and bandwidth utilization.
The LBL in the ESA-WF-FE-DIWGAN approach was significantly lower than that of the
previous methods (Table 5). ESA-WF-FE-DIWGAN took a variety of factors into account. ESA-WF-FE-DIWGAN has combined
TS-LB-FE-CNN-MPSO and TS-GOSH-FCE, the ESA-WF-FE-DIWGAN to split the resources, including
a decrease in the level of resource search that paves to effectual resource usage.
DIWGAN reaches greater accuracy in resource managing activities.
A decentralized computing system located amid data-producing devices and clouds is
known as fog computing. Users are able to allocate resources to enhance performance
to this flexible structure. Nevertheless, the application of novel virtualization
for task scheduling and resource management in fog-computing is hindered by a lack
of resources and low-latency services. Numerous studies have been conducted on scheduling
and LB in cloud computing. On the other hand, numerous LB initiatives have been presented
in the fog architectures. Some problems arise when considering how tasks are routed
between fog nodes and the cloud in different physical devices. Scheduling in fog is
highly challenging because of the volume and diversity of the devices. Only a few
studies have been done so far. LB is a particularly attractive and significant area
of study in FC because of its focus on maximizing resource utilization. LB has several
difficulties, including safety with fault tolerance. Table 6 shows that The Proposed ESA-WF-FE-DIWGAN model provided 4.52%, 7.67%, and 11.94%
lower make span than the existing methods, such as Talaat et al. [6], Tuli et al. [7], Teoh et al. [8]. The Proposed ESA-WF-FE-DIWGAN model provided 22.33%, 28.70%, and 14.79% lower cost
than existing methods like Talaat et al. [6], Tuli et al. [7], Teoh et al. [8] respectively. The proposed ESA-WF-FE-DIWGAN method provided26.07%, 14.36%, and 14.32%
lower LBL than the existing methods, such as Talaat et al. [6], Tuli et al. [7], Teoh et al. [8]. The ESA-WF-FE-DIWGAN method provided 9.80%, 14.58%, and 12.31% higher ARU than existing
methods, 2022, Talaat et al. [6], Tuli et al. [7], Teoh et al. [8].
Table 2. Comparison of Makespan.
|
Techniques
|
Makespan (ms)
|
|
50
|
90
|
130
|
150
|
|
TS-LB-FE-CNN-MPSO
|
34.22
|
44.34
|
63.33
|
34.33
|
|
TS-GOSH-FCE
|
56.45
|
45.77
|
67.56
|
66.33
|
|
ESA-WF-FE-DIWGAN (proposed)
|
23.32
|
15.67
|
5.33
|
23.20
|
Table 3. Comparison of Total cost.
|
Techniques
|
Total cost (%)
|
|
50
|
90
|
130
|
150
|
|
TS-LB-FE-CNN-MPSO
|
34.36
|
56.67
|
76.77
|
78.56
|
|
TS-GOSH-FCE
|
35.56
|
67.66
|
67.66
|
67.46
|
|
ESA-WF-FE-DIWGAN (proposed)
|
20.33
|
26.44
|
15.33
|
10.55
|
Table 4. Comparison of ARU.
|
Techniques
|
ARU(%)
|
|
50
|
90
|
130
|
150
|
|
TS-LB-FE-CNN-MPSO
|
76.77
|
65.45
|
79.77
|
82.45
|
|
TS-GOSH-FCE
|
80.34
|
82.34
|
75.34
|
69.66
|
|
ESA-WF-FE-DIWGAN (proposed)
|
91.34
|
90.77
|
92.34
|
95.66
|
Table 5. Comparison of LBL.
|
Techniques
|
LBL(%)
|
|
50
|
90
|
130
|
150
|
|
TS-LB-FE-CNN-MPSO
|
76.34
|
83.34
|
88.56
|
78.77
|
|
TS-GOSH-FCE
|
82.45
|
85.45
|
78.77
|
70.22
|
|
ESA-WF-FE-DIWGAN (proposed)
|
95.44
|
98.44
|
92.66
|
93.34
|
Table 6. Benchmark table.
|
Author name
|
Makespan (ms)
|
Total cost
|
LBL (%)
|
ARU (%)
|
|
Talaat et al. [6]
|
34.33
|
78.56
|
78.77
|
82.45
|
|
Tuli et al. [7]
|
66.33
|
67.46
|
70.22
|
69.66
|
|
Teoh et al. [8]
|
50.14
|
60.14
|
75.12
|
60.14
|
|
Sharma et al. [9]
|
-
|
-
|
-
|
-
|
|
ESA-WF-FE-DIWGAN (proposed)
|
23.20
|
10.55
|
93.34
|
10.55
|
Discussion
This paper proposed EDLB using DIWGAN. EDLB comprises three primary modules: (i) fog
resource monitoring, (ii) DIWGAN base classifier, (iii) Improved Dwarf Mongoose Optimization
algorithm. EDLB uses a dynamic real-time scheduling approach to achieve LB in fog
computing. FRM is accountable for tracking every server resource and protecting the
server data. The classifier depends on DIWGAN is accountable for categorizing every
fog server as appropriate or inappropriate. IDMO is accountable for choosing the best
server for the incoming process. EDLB decreases the response time and reaches higher
resource usage. This is a better mode to assure the uninterrupted service. Multiple
FC systems have been introduced in LB, but they all have numerous drawbacks. EDLB
deals with these drawbacks and achieves better performance in multiple scenarios.
Compared to other LB algorithms, EDLB achieves better make span, average resource
utilization, and load balancing levels. An unforeseen surge in demand on the server
responsible for that task may result in an EDLB real-time task failure. When the system
notices an abrupt increase in load on server hosting a task of real-time, it reschedules
a vital task for different server to address this problem.
4. Conclusion
DIWGANoptimized with an Improved Dwarf Mongoose Optimisation algorithmwas implemented
for scheduling the tasks (ESA-WF-FE-DIWGAN). The proposed ESA-WF-FE-DIWGAN approach
was executed by Java. The proposed ESA-WF-FE-DIWGAN approach achieved12.03%, which
was a 24.85% higher make span than existing methods, such as TS-LB-FE-CNN-MPSO and
TS-GOSH-FCE.
REFERENCES
S. Iftikhar, M.M.M. Ahmad, S. Tuli, D. Chowdhury, M. Xu, S.S. Gill, and S. Uhlig,
``HunterPlus: AI based energy-efficient task scheduling for cloud–fog computing environments,''
Internet of Things, vol. 21, p. 100667. Apr. 2023.
G. Shruthi, M.R. Mundada,, B.J Sowmya, and S. Supreeth, ``Mayfly tayloroptimisation-based
scheduling algorithm with deep reinforcement learning for dynamic scheduling in fog-cloud
computing,'' Applied Computational Intelligence and Soft Computing, vol. 198, no.
1, pp. 117273, Aug. 2022.
A. Pradhan, S.K. Bisoy, S Kautish, M.B. Jasser, and A.W. Mohamed, ``Intelligent Decision-Making
of Load Balancing Using Deep Reinforcement Learning and Parallel PSO in Cloud Environment,''
IEEE Access, vol. 10, no. 1, pp. 76939-76952. Jul. 2022.
Z. Movahedi, and B. Defude, ``An efficient population-based multi-objective task scheduling
approach in fog computing systems,'' Journal of Cloud Computing, vol. 10, no. 1, pp.
1-31. Dec. 2021.
M. Zahra, and D Bruno, A mohammad Hosseininia, ``An efficient population-based multi-objective
task scheduling approach in fog computing systems,'' Journal of Cloud Computing, vol.
10, no. 1, pp. 1-1, Dec. 2021.
F.M. Talaat, H.A. Ali, M.S. Saraya, and A.I. Saleh, ``Effective scheduling algorithmm
for load balancing in fog environment using CNN and MPSO,'' Knowledge and Information
Systems, vol. 64, no. 3, pp. 773-797. Mar. 2022.
S. Tuli, G. Casale, and N.R. Jennings, ``GOSH: Task scheduling using deep surrogate
models in fog computing environments,'' IEEE Transactions on Parallel and Distributed
Systems, vol. 33, no. 11, pp. 2821-2833. Dec. 2021.
Y.K.Teoh, S.S Gill, and A.K. Parlikad, ``IoT and fog computing based predictive maintenance
model for effective asset management in industry 4.0 using machine learning,'' IEEE
Internet of Things Journal. Vol. 10, no. 3, pp. 2087-2094, Jan. 2021.
O. Sharma, G. Rathee, and J. Kerrache Herrera-Tapia, ``Two-Stage Optimal Task Scheduling
for Smart Home Environment Using Fog Computing Infrastructures,'' Applied Sciences,
vol. 13, no. 5, p. 2939. C.A Feb. 2023.
Z. Shi, H. Li, Q. Cao, Z. Wang, and M. Cheng, ``A material decomposition method for
dual‐energy CT via dual interactive Wasserstein generative adversarial networks,''
Medical Physics, vol. 48, no. 6, pp. 2891-2905. Jun. 2021.
N. Krishnaraj, A. Daniel, K. Saini, and K. Bellam, ``EDGE/FOG computing paradigm:
Concept, platforms and toolchains''. In Advances in Computers (Vol. 127, no. 1, pp.
413-436). Elsevier, Jan. 2022.
T. Tuncer, S Dogan, and F. Ozyurt, ``An automated Residual Exemplar Local Binary Pattern
and iterative ReliefF based COVID-19 detection method using chest X-ray image,'' Chemometrics
and Intelligent Laboratory Systems, vol. 203, no. 1, pp. 104054. Aug. 2020.
J.O. Agushaka, A.E. Ezugwu, O.N. Olaide, O. Akinola, R.A. Zitar, and L. Abualigah,
``Improved Dwarf Mongoose Optimization for Constrained Engineering Design Problems,''
Journal of Bionic Engineering, vol. 1, no. 1, pp. 1-33. Mar. 2022.
Ravi Kumar Suggala, He is currently working as Assistant Professor, Department
of Information Technology, Shri Vishnu Engineering College for women, Bhimavaram,
Andhra Pradesh, India. His research interests include machine learning, deep learning,
and cloud technology.
Suma Bharathi. M, She is currently working as Assistant Professor, Department of
Information Technology, Shri Vishnu Engineering College for women, Bhimavaram, Andhra
Pradesh, India. Her research interests include Machine learning, deep learning, Grid
computing and soft computing.
P.L.V.D. Ravi Kumar, he is currently working as Assistant Professor, Department
of Information Technology, Shri Vishnu Engineering College for women, Bhimavaram,
Andhra Pradesh, India. His research interests include Machine learning, deep learning,
Grid computing and soft computing.
NVS Pavan Kumar Nidumolu is currently working as an Associate Professor in the
Department of Computer Science and Engineering at Koneru Lakshmiah Education Foundation,
Andhra Pradesh, India. He Completed his MCA degree from Andhra University, M.Tech
degree in Computer Science and Engineering from JNTUK and obtained his doctoral degree
in Department of Computer Science and Engineering. With over 20+ years of teaching
experience he is well known for his administrative role in various positions. He is
active and resourceful in various social bodies. He has over 15 of publication in
various national and international reputed journals. He has received many awards for
his achievements in administration. His area of interest is Machine learning and Data
Science.